Multi-domain data fusion step-by-step convergence community key crowd monitoring method and system

Abstract

The invention relates to the technical field of community monitoring, and discloses a multi-domain data fusion step-by-step convergence community key crowd monitoring method and system. Acquiring text data of social media, forums and news websites through an internet data acquisition module, preprocessing the text data, inputting the preprocessed text data into a large language model to generate high-dimensional word vectors, and identifying abnormal events by utilizing a clustering algorithm; mining an association rule of the abnormal event and government affair complaint data through a government affair data association module; marking a high-risk area and associated personnel information based on a space-time thermodynamic diagram analysis module; behavior characteristics and movement tracks of related personnel are captured through an articulated naturality web monitoring module; and finally, an analysis result is pushed to a grid member terminal for verification, and a monitoring strategy is dynamically updated. The system comprises an Internet data acquisition module, a text analysis module, a government affair association module, a space-time thermodynamic diagram analysis module, an articulated naturality web monitoring module and a grid member verification module. According to the invention, multi-source data fusion and intelligent analysis are realized, and the accuracy and timeliness of monitoring key crowds in a community are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of community monitoring, and in particular to a method and system for monitoring key groups in a community with multi-domain data fusion and step-by-step convergence. Background Art

[0002] With the rapid development of society, community management is facing more and more complex crowd monitoring tasks, especially the monitoring and management of key groups such as the unemployed, poor families, migrant population, patients with severe mental disorders, community correction personnel, released prisoners and drug-related persons. At present, the monitoring of key groups in the community mainly relies on traditional video surveillance systems or simple government data statistics, but this single data source method has serious limitations.

[0003] First, traditional monitoring methods cannot fully reflect the behavioral characteristics and potential risks of key groups, and lack the effective use of Internet information, making it difficult to capture the attention and discussion of key groups outside the community. Secondly, the existing monitoring system has a low level of intelligence and cannot automatically identify abnormal events, requiring a lot of human intervention. Although some communities have tried data fusion technology, it still remains at the simple data superposition stage and fails to achieve in-depth analysis and dynamic monitoring. More importantly, the existing methods lack a dynamic tracking mechanism for the behavior of key groups, do not fully utilize the analytical capabilities of large language models, and are difficult to cope with complex community management challenges. Summary of the invention

[0004] The present invention aims to provide a community key population monitoring system and method based on multi-domain data fusion and step-by-step convergence, so as to solve the problems of single data source, low intelligence and lack of dynamic feedback in the prior art, break through the limitations of traditional monitoring technology, realize accurate identification and dynamic management of key populations, and significantly improve the level of monitoring intelligence and decision-making reliability.

[0005] According to the design scheme provided by the present invention, on the one hand, a community key population monitoring method with multi-domain data fusion and step-by-step convergence is provided, comprising:

[0006] Through Internet data collection and monitoring, we obtain community-related text data from social media, forums, and news websites, and perform preprocessing on the collected text data by cleaning, segmenting, removing stop words, and extracting stems to generate preprocessed text sequences;

[0007] The preprocessed text sequence is input into the large language model to generate context-related high-dimensional word vector representations. The word vectors are clustered and analyzed using a clustering algorithm to identify and mark abnormal events that deviate from normal behavior patterns and extract key information about the abnormal events.

[0008] Obtain community-related complaints, suggestions and event data through the government network platform, conduct correlation analysis between government data and abnormal event information identified on the Internet through association rule mining algorithms, and screen out complaints and suggestions data related to abnormal events whose support and confidence are higher than the preset threshold;

[0009] Calculate the spatiotemporal density distribution of abnormal events based on the spatiotemporal heat map analysis method, mark high-risk areas where event density is higher than the preset threshold and their corresponding event types and related personnel information, and screen out clues involving key groups;

[0010] Through the visual network of intelligent cameras and sensor networks, the target detection model is used to capture the behavioral characteristics and activity trajectory of the marked relevant personnel, generate a monitoring report containing behavioral characteristics and trajectories, and verify and supplement the information of key groups;

[0011] The results of Internet text analysis, government data correlation analysis and monitoring reports are pushed to the community grid worker terminals to guide on-site verification, and the monitoring strategy is dynamically updated through the feedback loop formula based on the verification feedback results to achieve step-by-step convergence and accurate identification of monitoring targets.

[0012] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the Internet data collection is realized by crawler technology, and regular expressions are used to clean the original data to remove HTML tags and advertising content.

[0013] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the large language model is a pre-trained model based on the Transformer architecture, and the process of generating word vectors is expressed as:

[0014]

[0015] in, is the preprocessed text sequence, is the output context-dependent word vector.

[0016] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the clustering algorithm is a K-means algorithm, and its objective function is expressed as:

[0017]

[0018] in, is the number of cluster categories, is the cluster centroid.

[0019] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the association rule mining adopts Apriori algorithm, and the confidence calculation is expressed as:

[0020]

[0021] in, , For a project data item set, the support threshold is 0.1 and the confidence threshold is 0.5.

[0022] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the spatiotemporal heat map analysis method calculates the event density as:

[0023]

[0024] in, Represents the time dimension, represents the spatial dimension, is the smoothing parameter.

[0025] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the behavior verification is implemented by the YOLOv5 model, and its output is:

[0026]

[0027] in, Represents the input image.

[0028] As a community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention, further, the feedback loop formula is:

[0029]

[0030] in, represents the feedback weight matrix, is the current monitoring status. Indicates new information obtained by grid workers during their visits.

[0031] In another aspect, the present invention provides a community key population monitoring system for multi-domain data fusion and step-by-step convergence, which is used to implement the above method, including:

[0032] Internet data collection module, which is used to obtain community-related text data from social media, forums and news websites through crawler technology, and pre-process the data by cleaning, segmenting and stemming;

[0033] Text analysis and anomaly detection module, including large language models and clustering algorithm units, used to generate high-dimensional word vectors and identify abnormal events that deviate from normal behavior patterns;

[0034] The government data association module is used to access government platform data, associate government data with Internet abnormal event information through association rule mining algorithms, and screen complaints and suggestions data with support and confidence levels higher than preset thresholds;

[0035] The spatiotemporal heat map analysis module is used to calculate the spatiotemporal density distribution of abnormal events, mark high-risk areas and related personnel information;

[0036] The visual network monitoring module includes smart cameras and target detection models deployed in the community to capture the behavioral characteristics and activity trajectories of relevant personnel and generate monitoring reports;

[0037] The grid worker verification and feedback module is used to receive Internet analysis results, government-related results and monitoring reports, support on-site verification and dynamically update monitoring strategies through feedback loop formulas;

[0038] Data storage and processing server, used to store multi-source data and perform data fusion and computing tasks of each module.

[0039] The beneficial effects of the present invention are:

[0040] The method and system for monitoring key groups in the community with multi-domain data fusion and step-by-step convergence proposed in the present invention have achieved multiple technological innovations in the field of intelligent community governance and have significant application value. At the data integration level, an innovative three-dimensional fusion architecture of Internet text data, government platform data, and visual network monitoring data is constructed. Through the cross-validation and complementarity of multi-source heterogeneous data, the data island problem in the traditional monitoring system is effectively solved, making the analysis of the behavioral characteristics of key groups more comprehensive and accurate. In terms of information processing mechanism, a step-by-step convergent progressive processing flow from data collection, text analysis, government affairs association to visual network verification is designed. Through four-layer processing and screening steps, the intelligent convergence of monitoring information is realized, which significantly improves the monitoring accuracy.

[0041] In terms of intelligent applications, the system deeply integrates natural language processing and computer vision technologies. Through the coordinated application of the semantic understanding ability of large language models and target detection technology, it can efficiently identify abnormal behavior patterns in texts and images, greatly improving the timeliness of abnormal event discovery. At the same time, the solution has built a complete dynamic optimization mechanism. Through the continuous interaction between the on-site verification of grid members and the algorithm model, it realizes the adaptive adjustment of monitoring strategies and forms a collaborative optimization closed loop of "machine intelligence + manual experience". It not only improves the modernization level of community governance, but also provides a reusable technical solution for the construction of smart cities, which has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of the community key population monitoring method with multi-domain data fusion and step-by-step convergence of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] Embodiment 1:

[0045] This embodiment provides a method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence. Figure 1 As shown, the specific implementation steps of the method include:

[0046] S1. Obtain community-related text data from social media, forums, and news websites through Internet data collection and monitoring, perform preprocessing on the collected text data by cleaning, word segmentation, stop word removal, and stem extraction, and generate preprocessed text sequences;

[0047] S2. Input the preprocessed text sequence into the large language model to generate context-related high-dimensional word vector representation, perform cluster analysis on the word vector through the clustering algorithm, identify and mark abnormal events that deviate from normal behavior patterns, and extract key information of abnormal events;

[0048] S3. Obtain community-related complaints, suggestions and event data through the government network platform, conduct correlation analysis between government data and abnormal event information identified on the Internet through association rule mining algorithms, and screen out complaints and suggestions data related to abnormal events whose support and confidence are higher than the preset threshold;

[0049] S4. Calculate the spatiotemporal density distribution of abnormal events based on the spatiotemporal heat map analysis method, mark high-risk areas where the event density is higher than the preset threshold and their corresponding event types and relevant personnel information, and screen out clues involving key groups;

[0050] S5. Through the visual network smart camera and sensor network, use the target detection model to capture the behavioral characteristics and activity trajectory of the marked relevant personnel, generate a monitoring report containing behavioral characteristics and trajectories, and verify and supplement the information of key groups;

[0051] S6. Push the Internet text analysis results, government data correlation analysis results and monitoring reports to the community grid worker terminal to guide on-site verification, and dynamically update the monitoring strategy through the feedback loop formula based on the verification feedback results to achieve step-by-step convergence and accurate identification of monitoring targets.

[0052] In the specific implementation, by deploying a distributed crawler system, community-related text data from social media (such as Weibo, WeChat public platform), community forums (such as Tieba) and news websites (such as People's Daily Online, local news websites) are collected in real time. The crawler system uses dynamic IP proxy technology to circumvent the anti-crawling mechanism to ensure the continuity and stability of data collection. For the collected raw data, regular expressions are first used to remove HTML tags, advertising content and special characters, and then the HanLP word segmentation tool is used for Chinese word segmentation processing, and irrelevant words are filtered out in combination with the stop word list. Finally, the PorterStemmer algorithm is used to extract stems and generate standardized text sequences. The preprocessed text sequence is represented in vector form:

[0053]

[0054] in, Represents the i-th word or subword unit.

[0055] The text sequence is input into a pre-trained large language model based on the Transformer architecture (such as BERT or GPT-3). The model generates context-dependent high-dimensional word vectors through a multi-layer self-attention mechanism:

[0056]

[0057] in, Represents the input text sequence, It is a context vector representation generated by the Transformer encoder of LLM, which contains the semantic information of the behavior sequence.

[0058] Use the K-means clustering algorithm to perform unsupervised clustering on word vectors and set the number of clusters. The value is adjusted dynamically according to the silhouette coefficient, the Euclidean distance between the sample and the centroid is calculated, and abnormal clusters that deviate from the main cluster center are identified:

[0059]

[0060] Extract high-frequency keywords (such as "gathering", "fighting", "abnormal behavior") in the cluster and the associated spatiotemporal information, and mark them as potential abnormal events.

[0061] Obtain community complaints, suggestions and event data regularly through the government network API interface (such as the national government service platform open interface), with fields including event type, occurrence time, geographic location, processing status, etc. Use the Apriori algorithm to mine the association rules between Internet abnormal events and government data, set the minimum support threshold of 0.1 and the confidence threshold of 0.5, and the calculation rules:

[0062]

[0063] Among them, X and Y are project data item sets, and strongly related events (such as "night noise complaints" and "gathering events discussed in forums") are screened out.

[0064] Based on the correlation results, the Gaussian kernel density estimation method is used to calculate the spatiotemporal distribution of events:

[0065]

[0066] Among them, t represents the time distribution, l represents the location distribution, It is a smoothing parameter with a default value of 0.5 (which can be adjusted according to the design scale). The areas with marker density higher than the threshold (such as Top10%) are high-risk areas, which are associated with the event type (such as public security incidents, activities of patients with mental disorders) and the ID of the person involved.

[0067] Smart cameras and LoRa sensor networks deployed in high-risk areas of the community collect video stream data in real time. Detect target persons in the video using the YOLOv5 model:

[0068]

[0069] Where I represents the input image, and the output includes the coordinates of the person's bounding box, behavior labels (such as "wandering" and "running"), and trajectory time series data. The detection results are matched with the person ID in the government data to generate a structured monitoring report, including behavior frequency, activity hotspots, and abnormal behavior scores.

[0070] The analysis results are pushed to the grid worker's mobile terminal (such as a customized APP), prompting the list of people to be verified, event details and recommended verification time. After the grid worker visits the site, he submits the verification results (such as "true" or "false alarm") and supplementary information (such as new personnel relationships) through the APP. The system dynamically updates the monitoring strategy based on feedback:

[0071]

[0072] in, represents the feedback weight matrix, is the current monitoring status. Indicates new information obtained by grid workers during their visits.

[0073] Embodiment 2:

[0074] This embodiment provides a community key population monitoring system for multi-domain data fusion and step-by-step convergence, which is used to implement the above method, including:

[0075] Internet data collection module, which is used to obtain community-related text data from social media, forums and news websites through crawler technology, and pre-process the data by cleaning, segmenting and stemming;

[0076] Text analysis and anomaly detection module, including large language models and clustering algorithm units, used to generate high-dimensional word vectors and identify abnormal events that deviate from normal behavior patterns;

[0077] The government data association module is used to access government platform data, associate government data with Internet abnormal event information through association rule mining algorithms, and screen complaints and suggestions data with support and confidence levels higher than preset thresholds;

[0078] The spatiotemporal heat map analysis module is used to calculate the spatiotemporal density distribution of abnormal events, mark high-risk areas and related personnel information;

[0079] The visual network monitoring module includes smart cameras and target detection models deployed in the community to capture the behavioral characteristics and activity trajectories of relevant personnel and generate monitoring reports;

[0080] The grid worker verification and feedback module is used to receive Internet analysis results, government-related results and monitoring reports, support on-site verification and dynamically update monitoring strategies through feedback loop formulas;

[0081] Data storage and processing server, used to store multi-source data and perform data fusion and computing tasks of each module.

Claims

1. A method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence, characterized in that: include: Through Internet data collection and monitoring, we obtain community-related text data from social media, forums, and news websites, and perform preprocessing on the collected text data by cleaning, segmenting, removing stop words, and extracting stems to generate preprocessed text sequences; The preprocessed text sequence is input into the large language model to generate context-related high-dimensional word vector representations. The word vectors are clustered and analyzed using a clustering algorithm to identify and mark abnormal events that deviate from normal behavior patterns and extract key information about the abnormal events. Obtain community-related complaints, suggestions and event data through the government network platform, conduct correlation analysis between government data and abnormal event information identified on the Internet through association rule mining algorithms, and screen out complaints and suggestions data related to abnormal events whose support and confidence are higher than the preset threshold; Calculate the spatiotemporal density distribution of abnormal events based on the spatiotemporal heat map analysis method, mark high-risk areas where event density is higher than the preset threshold and their corresponding event types and related personnel information, and screen out clues involving key groups; Through the visual network of intelligent cameras and sensor networks, the target detection model is used to capture the behavioral characteristics and activity trajectory of the marked relevant personnel, generate a monitoring report containing behavioral characteristics and trajectories, and verify and supplement the information of key groups; The results of Internet text analysis, government data correlation analysis and monitoring reports are pushed to the community grid worker terminals to guide on-site verification, and the monitoring strategy is dynamically updated through the feedback loop formula based on the verification feedback results to achieve step-by-step convergence and accurate identification of monitoring targets.

2. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The Internet data collection is achieved through crawler technology, and regular expressions are used to clean the original data to remove HTML tags and advertising content.

3. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The large language model is a pre-trained model based on the Transformer architecture. The process of generating word vectors is expressed as follows: ； in, is the preprocessed text sequence, is the output context-dependent word vector.

4. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The clustering algorithm is the K-means algorithm, and its objective function is expressed as: ； in, is the number of cluster categories, is the cluster centroid.

5. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The association rule mining adopts the Apriori algorithm, and the confidence calculation is expressed as: ； in, , For a project data item set, the support threshold is 0.1 and the confidence threshold is 0.

5.

6. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The spatiotemporal heat map analysis method calculates the event density as: ； in, Represents the time dimension, represents the spatial dimension, is the smoothing parameter.

7. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The behavior verification is implemented by the YOLOv5 model, and its output is: ； in, Represents the input image.

8. The method for monitoring key populations in a community with multi-domain data fusion and step-by-step convergence according to claim 1 is characterized in that: The feedback loop formula is: ； in, represents the feedback weight matrix, is the current monitoring status. Indicates new information obtained by grid workers during their visits.

9. A community key population monitoring system with multi-domain data fusion and step-by-step convergence, characterized in that: Used to implement the method according to any one of claims 1 to 8, comprising: Internet data collection module, which is used to obtain community-related text data from social media, forums and news websites through crawler technology, and pre-process the data by cleaning, segmenting and stemming; Text analysis and anomaly detection module, including large language models and clustering algorithm units, used to generate high-dimensional word vectors and identify abnormal events that deviate from normal behavior patterns; The government data association module is used to access government platform data, associate government data with Internet abnormal event information through association rule mining algorithms, and screen complaints and suggestions data with support and confidence levels higher than preset thresholds; The spatiotemporal heat map analysis module is used to calculate the spatiotemporal density distribution of abnormal events, mark high-risk areas and related personnel information; The visual network monitoring module includes smart cameras and target detection models deployed in the community to capture the behavioral characteristics and activity trajectories of relevant personnel and generate monitoring reports; The grid worker verification and feedback module is used to receive Internet analysis results, government-related results and monitoring reports, support on-site verification and dynamically update monitoring strategies through feedback loop formulas; Data storage and processing server, used to store multi-source data and perform data fusion and computing tasks of each module.

Images